ZetaForge is an AI platform for rapid development and deployment of advanced safe AI solutions. Easilly assemble reusable, customizable and containerized blocks into highly visual AI pipelines.
TBD
1- Install Docker
Get Docker here
2- Verify Docker is correctly installed
docker --version
3- Install the Python packages
pip install -r requirements.txt
4- cd to \backend\nodejs and install the packages
npm install
5- cd to \frontend and install the packages
npm install
Use the following command to launch Forge in your localhost.
cd to \backend\nodejs
node server.js
Examples of Pipeline of Block are provided in the pipelines folder. They can be loaded and run using the Load and Run top menu buttons.
Zetaforge demo and code overview https://drive.google.com/drive/folders/1oeX6Sd8sic1cw4WSUk_JJ7jDuMhKvNqY?usp=sharing
Each block (Python) is compose of a computations.py, requirements.txt, Dockerfile and specs.json file which follows the templates below. In a block folder wrapping up a full ML repo, there can be any nescessary files as long the compute function from computations.py follows the interface below.
Specs.json files will have the same template for all programming languages and blocks will follow a similar layout.
specs.json
{
"block": {
"information": {
"id": "minimal",
"name": "Minimal Block",
"description": "description",
"system_versions": [],
"block_version": "block version number"
},
"inputs": {
"in1": {
"type": "int",
"connections": [],
"relay_connections": []
},
"in2": {
"type": "str",
"connections": [],
"relay_connections": []
}
},
"outputs": {
"out1": {
"type": "int",
"connections": [],
"relay_connections": []
},
"out2": {
"type": "str",
"connections": [],
"relay_connections": []
}
},
"parameters": {},
"action": {
"container_uuid": "minimal",
"container_image_uuid": "minimal",
"block_source": "blocks/minimal",
"version:": "latest"
},
"views":{
"mode": "modal",
"node": {
"active": "True or False",
"title_bar": {
"background_color": "#D55908"
},
"preview": {
"active": "false"
},
"html": "",
"pos_x": "300",
"pos_y": "200",
"pos_z": "999"
}
},
"events": []
}
}
Note that for preview to display when set to true, it need to be a view block that has an html output variable.
computations.py
# Python block template
compute(in1, in2, in3,...):
'''A textual description of the compute function.
Inputs:
in1 (all): Textual description of in1
in2 (all): Textual description of in2
in3 (all): Textual description of in2
Outputs:
out1 (all): Textual description of out1
out2 (all): Textual description of out2
'''
#Your code
return {'out1': out1, 'out2': out2, ...}
To create a new block, drag and drop the New_Python sidebar library block. Insert your code or ask GPT to generate code based on your prompt. Please, ensure you follow the provided template. After you save the block a new block will be created in the folder my_blocks. Using the Load button, you can load previously created blocks by selecting the block's specs.json file.
You can also copy the folder blocks/minimal into the my_blocks folder and manually edit files.
In manager.py there is a flag save_graph_to_database = False. To record each run in the database and access the result events from each block set it to True. In index.html, you can comment and uncomment the history sections.
You can call your PostgreSQL database graph_db1 and use credentials like in manager.py after the save_graph_to_database flag.
In your database, make a table called my_table with columns id (serial, primary key) and data (json)
See Zetane Docs (TBD)
From the folder backend/python, the following command with launch only the pipeline computing for testing and debugging.
python ../python/manager.py
TBD
- Need to add a stop button to cancel computations and reset the server
- blocks to blocks
- Reset colors when rerunning graph
- Versioning
- Fix index.html console errors
- Add font awesome icons in specs.json
- Test Kubernetes
TBD
TBD
Version 0.1 (internal alpha)